Problem with push_to_hub

🤗Transformers
1

Hello everyone. I’m trying to upload my fine-tuned GPT-2 Model to Model Hub.
When I try to use the uploading function push_to_hub I get the following error:
AttributeError: 'GPT2Model' object has no attribute 'push_to_hub'
In the documentation it says that I can push the model with this function.
Can anybody help please?

2

You need to update your Transformers library to the latest version.

1 Like
4

Thank you so much. I have another question. Is there a way to push a model to the model hub after training without saving the new model to the disk with trainer.save_model() ? Would Trainer.push_to_hub do that or is it for pushing just the trainer instance? If yes how can I give my authentication token as a parameter to trainer.push_to_hub method?

5

You can set your token in the training arguments inside push_to_hub_token.
The Trainer will save the model on disk with save_model, as you need to save it to the repo of the model you will be using before it can git push it to the model hub.

1 Like
6

So you mean I should set push_to_hub_token in Training arguments. Then train the model and save it using trainer.save_model() function. Then I should just recreate it from the saved file using

fine_tuned_model = AutoModel.from_pretrained("./file_path")

And then finally push it to hub using

fine_tuned_model.push_to_hub("Name of model")
7

No, you should just do trainer.push_to_hub() once you’re done with training. Look at any of the notebooks examples for more context, or the push to hub video.

2 Likes
8

Thank you very much. I was able to solve my problem after I saw the training argument “push_to_hub_model_id” and transformers-cli from your example notebook “Train a language model”.

1 Like
9

‘Functional’ object has no attribute ‘push_to_hub’

screenshot-colab.research.google.com-2023.01.28-12_24_29
screenshot-colab.research.google.com-2023.01.28-12_24_291844×658 79 KB

I am facing this issue regarding push_to_hub API which does not support Functional models from TensorFlow. My model is an NMT that has custom objects.
It is a custom model which I have built from scratch and does not use any methods like from_pretrained.
I want to push it to huggingface as well as create a pipeline

Please help